Re: Design question
- From: "Vadym Stetsyak" <vadym_s@xxxxxxx>
- Date: Mon, 23 Jan 2006 17:50:57 +0200
> In fact, the start phase is not that difficult to handle.
> The problem is that, for now, when a connection between 2 hosts is
> broken, a message is sent (via TCP connection) to every other hosts
> connections to keep each hosts table uptodate and an automatic re-
> connection between these 2 hosts is tried triggered by a timer.
doesn't this introduce inefficiency? You'll get implementation complexity,
synchronization lag,
and ineffiecient network usage ( additional network traffic for table sync )
for now you have 60 connections on each host, connection is broken between 2
of them.
You have 60 machine tables, which can become desynchronized for considerable
timeout.
With server-based approach you will have 1 ( several ) machine tables, which
will be synchronized.
Consider also traffic through the network, when machines need to synchronize
machine table.
For failover you can introduce 2 or more servers: primary and secondary.
Hosts will know about the second one, but will work with the 1-st. If
primary server is down, they will automatically reconnect to the secod one.
Primary and secondary servers will have synchronized machine table.
--
Vadym Stetsyak aka Vadmyst
http://vadmyst.blogspot.com
.
- Follow-Ups:
- Re: Design question
- From: Droopy
- Re: Design question
- References:
- Design question
- From: Droopy
- Re: Design question
- From: Vadym Stetsyak
- Re: Design question
- From: Droopy
- Re: Design question
- From: Vadym Stetsyak
- Re: Design question
- From: Droopy
- Design question
- Prev by Date: Re: Clarification please
- Next by Date: UDP's sendto() from two IP Addresses to IP Address from small LAN
- Previous by thread: Re: Design question
- Next by thread: Re: Design question
- Index(es):
Relevant Pages
|